Accuracy of atomization energies and reaction enthalpies in standard and extrapolated electronic wave function/basis set calculations

Abstract

The accuracy of standard ab initio wave-function calculations of atomization energies and reaction enthalpies has been assessed by comparing with experimental data for 16 small closed-shell molecules and 13 isogyric reactions. The investigated wave-function models are Hartree–Fock (HF), Møller–Plesset second-order perturbation theory (MP2), coupled-cluster theory with singles and doubles excitations (CCSD) and CCSD with perturbative triple-excitation corrections [CCSD(T)]; the one-electron basis sets used are the correlation-consistent cc-pVxZ and cc-pCVxZ basis sets with cardinal numbers x=D, T, Q, 5, and 6. Results close to the basis-set limit have been obtained by using two-point extrapolations. In agreement with previous studies, it is found that the intrinsic error of the CCSD(T) method is less than chemical accuracy (≈4 kJ/mol) for both atomization energies and reaction enthalpies. The mean and maximum absolute errors of the best CCSD(T) calculations are 0.8 and 2.3 kJ/mol for the atomization energies and 1.0 and 2.3 kJ/mol for the reaction enthalpies. Chemical accuracy is obtained already from the extrapolations based on the cc-pCVTZ and cc-pCVQZ basis sets—with mean and maximum absolute errors of 1.7 and 4.0 kJ/mol for atomization energies and 1.3 and 3.1 kJ/mol for reaction enthalpies. The intrinsic errors of the Hartree–Fock, MP2, and CCSD wave-function models are significantly larger than for CCSD(T). For CCSD and MP2, the mean absolute errors in the basis set limit are about 32 kJ/mol for the atomization energies and about 10 and 15 kJ/mol, respectively, for the reaction enthalpies. For the Hartree–Fock model, the mean absolute errors are 405 and 29 kJ/mol for atomization energies and reaction enthalpies, respectively. Correlation of the core electrons is important in order to obtain accurate results with CCSD(T). Without compromising the accuracy, the core contribution may be calculated with a basis set that has one cardinal number lower than that used for the valence correlation contribution. Basis-set extrapolation should be used for both the core and the valence contributions.